Re: [PATCH rdma-next v3 3/6] RDMA/irdma: Support UMEM attribute for QP creation

[Leon Romanovsky <leon@kernel.org>]

On Tue, Oct 27, 2026 at 06:37:56PM -0400, David Hu wrote:
> On Mon, Oct 26, 2026 at 8:02 AM Jacob Moroni <jmoroni@google.com> wrote:
> >
> > Hi David,
> >
> > Looks good in general, but since this IRDMA driver is used with iWARP
> > NICs today,
> > can you double check this Sashiko finding? Maybe this newer path can be rejected
> > for iWARP NIC types at first? Or if there's a way to make this new path work for
> > iWARP, even better. I am not sure if there's some generic way to map the generic
> > umem for CPU access that works even for dmabufs?
> 
> Hi Jacob,
> 
> Thank you for the feedback. I double-checked the Sashiko finding -
> yes, we can make the rejection explicit. I'll send out a v4 shortly.
> 
> re: Making the path work for dma-buf + iWARP.  This seems doable; it
> just requires moderate work to plumb iWARP specific quirks to be
> dma-buf aware. However, I wasn't able to find a compelling real-world
> use case for this. If there is a strong preference to pursue this as
> part of this series, please let me know.

I wouldn't recommend investing too much effort in this. Ultimately, iWARP is
not widely used in practice, so dma-buf + iWARP is unlikely to see much use.

Thanks